THE STATISTICAL AND GEOGRAPHICAL ANALYSIS ON THE IMPACTS OF SOCIOECONOMIC CHARACTERISTICS ON BUS-STOP DAILY BOARDING IN RICHMOND CITY

Transcription

1 Virginia Commonwealth University VCU Scholars Compass Theses and Dissertations Graduate School 2015 THE STATISTICAL AND GEOGRAPHICAL ANALYSIS ON THE IMPACTS OF SOCIOECONOMIC CHARACTERISTICS ON BUS-STOP DAILY BOARDING IN RICHMOND CITY Yue Zhao Follow this and additional works at: The Author Downloaded from This Thesis is brought to you for free and open access by the Graduate School at VCU Scholars Compass. It has been accepted for inclusion in Theses and Dissertations by an authorized administrator of VCU Scholars Compass. For more information, please contact

2 Yue Zhao 2015 All Rights Reserve

3 The Statistical and Geographical Analysis on the Impacts of Socioeconomic Characteristics on Bus-Stop Daily Boarding in Richmond City A thesis submitted in partial fulfillment of the requirements for the degree of Master of Urban and Regional Planning at Virginia Commonwealth University By Yue Zhao Master of Urban and Regional Planning Major Advisor: Xueming Chen, Ph.D, Associate Professor Master of Urban and Regional Planning Program Virginia Commonwealth University L. Douglas Wilder School of Government and Public Affairs Richmond, Virginia December, 2015

4 ii Acknowledgement I would like to thank my thesis major advisor Dr. Chen. I could not have accomplished this thesis without his help. This thesis could not have been completed without the time and energy spent by Dr. Suen and Dr. Yan, whose ideas and suggestions are helpful and constructive. And I sincerely appreciate my parents who have educated and supported me so much.

5 iii Table of Contents Acknowledegment... ii List of figures...v List of tables..vii Abstract.....viii Chapter 1: Introduction 1.1 Existing Problems Research Objectives Research Methods Research Scope Research Contribution Research Factors 4 Chapter 2: Literature review 2.1 The public transit situation and issues of the U.S and Richmond, VA Previous study of factors impacting transit ridership Internal and external factors.16 Chapter 3: Methodology 3.1 Method and data collection Why you use these methods Define variables Hypothesis 29

6 iv Chapter 4: Analysis 4.1 Mapping dependent and independent variables Statistics Model analysis Model Validation.45 Chapter 5: Conclusion 49 APPENDIX A: CORRELATION MODEL SUMMARY (All Variables) 51 APPENDIX B: CORRELATION MODEL SUMMARY (All Variables without Total Population and Acres) 54 APPENDIX C: CORRELATION MODEL SUMMARY (All Variables without Household Density, Population Density and Employment Density) 57 APPENDIX D: THE DATA OF 161 BLOCK GROUPS.60 References.. 65

7 v List of Figures Figure 1: Percentage of Total Ridership by Route Category (Source: GRTC GFI DATA).3 Figure 2: Annual transit ridership and annual ridership per capita in the US [Source: American Public Transportation Association (2001)].6 Figure 3: Annual Transit Ridership. [Source: American Public Transportation Association (APTA)] Figure 4: Transit Mode Share. [Source: American Public Transportation Association (APTA)]...8 Figure 5: Richmond Greater Richmond Transit Company (GRTC) Bus Route Map. Comprehensive Operations Analysis (2008) (Source: GRTC) 9 Figure 6: Bus Service Areas and Population Density. Comprehensive Operations Analysis (2008) (Source: GRTC)..10 Figure 7: Combined Ranking of Access to Transit and Employment, 100 Metropolitan Areas (Source: Tomer et al., 2011)..11 Figure 8: Studies of Transit Ridership. [Source: (Taylor and Fink, 2003)]..15 Figure 9: Daily-On.32 Figure 10: Total population...32 Figure 11: Income..33 Figure 12: Vehicles 33 Figure 13: Population Density...34

8 vi Figure 14: Household Density...34 Figure 15: Black Ratio...35 Figure 16: Acres.35 Figure 17: Bus line count...36 Figure 18: Female Rate.36 Figure 19: Low Education Population...37 Figure 20: Households...37 Figure 21: Employment Density 38 Figure 22: Low-Skilled Employment 38 Figure 23: Unemployment Rate.39 Figure 24: Percent Error by Block Group in Richmond City 47

9 vii List of Tables Table 1: Direct Strategies. Direct and Indirect Strategies of Transit Ridership Project in the Study by European Commission Transportation Research (1996). (Source: European Commission Transportation Research).. 13 Table 2: Indirect Strategies. Direct and Indirect Strategies of Transit Ridership Project in the Study by European Commission Transportation Research (1996). (Source: European Commission Transportation Research)..14 Table 3: Regression Model Parameter Estimates in Production Side (Source: (Chen and Suen, 2010))...23 Table 4: Regression Model Parameter Estimates in Attraction Side (Source: (Chen and Suen, 2010))...24 Table 5: Dependent and Independent Variables 28 Table 6: Hypothesis of independent factors...30 Table 7: Regression model summary and parameter estimates.40 Table 8: Regression Model Parameter Estimates..42 Table 9: Summary of key impacting variables..44

10 Abstract THE STATISTICAL AND GEOGRAPHICAL ANALYSIS ON THE IMPACTS OF SOCIOECONOMIC CHARACTERISTICS ON BUS-STOP DAILY BOARDING IN RICHMOND CITY By Yue Zhao, Master of Urban and Regional Planning A thesis submitted in partial fulfillment of the requirements for the degree of Master of Urban and Regional Planning at Virginia Commonwealth University Virginia Commonwealth University, 2015 Major Advisor: Xueming Chen, Ph.D, Associate Professor, Master of Urban and Regional Planning Program At present, Richmond, Virginia only has bus transit services provided by the Greater Richmond Transit Company (GRTC) and primarily concentrated within the boundary of Richmond City. GRTC is impacted by both supply-side and demand-side factors, notably socioeconomic characteristics of bus riders, bus ridership is unevenly distributed across different bus stops. This thesis will conduct statistical and geographical analysis on the impacts of socioeconomic characteristics on bus-stop daily boarding in Richmond City. The statistical analysis includes both correlation analysis and regression analysis, assuming one dependent variable (bus-stop daily boarding) and fourteen independent variables (most of which describe socioeconomic characteristics of bus riders) at aggregated census

11 block group levels. The research concentrates on local bus routes and the block groups with local bus stops in Richmond. This empirical study aims to identify the significant factors impacting bus ridership and assess the bus service situation for affected block groups (under-served or over-served). The study outcomes, such as the number of bus lines as the most important factor impacting ridership, will have important implications for Richmond s local transit planning and decision-making.

12 CHAPTER 1: INTRODUCTION 1.1 Existing Problems The Greater Richmond Transit Company (GRTC) currently operates a hub-and-spoke bus transit system for the Richmond region. GRTC s local service area includes most portions of the Richmond City, significant parts of Henrico County, and limited areas of Chesterfield County. GRTC s express routes, which are not studied in this thesis due to its diffused locations and the nature of express service, serve to offer commuting choices rather than provide a comprehensive transit service. The stop-level bus ridership of GRTC s local service is unevenly distributed in the Richmond region. Examining its cause motivates this study, which should have important planning and policy implications for GRTC and local community. 1.2 Research Objectives This thesis has three objectives: To develop and evaluate bus-stop level ridership models using GIS and statistical methods to identify the significant factors which impact public transit ridership. 1

13 To identify through the ridership modeling analysis, block groups are under-served or over-served by GRTC bus lines. To provide important policy implications on GRTC s transportation decision-making and local transportation development. 1.3 Research Methods This study is based on literature reviews and a rigorous data analysis on the transit ridership influenced by block group-level socioeconomic variables. Geographic information system (GIS) is used to show different transit ridership situations by bus stop service and regression analysis to examine and analyze factors which may significantly influence public transit ridership. 1.4 Research Scope My research concentrates on local bus routes and the block groups which have local bus stops in Richmond City. GRTC s bus route structure is classified as a hub-and-spoke system, where service converges on a central downtown area and then spreads out into the surrounding neighborhoods. But according to the 2007 household survey in Comprehensive Operations Analysis conducted by GRTC (2008), 86 percent of fixed route service is provided to the downtown area, most local bus routes support downtown 2

14 area, and only express routes provide direct service between downtown and suburbs. According to the GRTC Annual Ridership in calendar year 2010, more than 80 percent of the ridership was made by local bus routes. (Figure 1) Figure 1: Percentage of Total Ridership by Route Category (Source: GRTC GFI DATA, 2010) 1.5 Research Contribution Researching and understanding the important factors that influence public transit ridership and existing deficiencies are useful to support a sustainable transit system that reduces traffic congestion, provides mobility to the disadvantaged, enhances economic development, conserves energy and improve air quality. Assessing each affected block group s ridership supports that goal and offers information valuable for policies established by Richmond s transit and planning agencies. 3

15 1.6 Research Factors Population (total population), population density (total persons/acre),employment density (employees/acre), income (median household income) car ownership (the number of vehicles in household), acres (the size of each block group), black ratio (black population/total population), female rate (total female population/ total population),households (total households), household density (households/acre), lowskilled jobs (total low-skilled employment), low educated people (educational attainment (under and not including high school)), unemployment rate (total unemployed/ total population), and the number of bus lines in each block group are my independent variables. The daily number of passengers boarding on a bus at any bus stop aggregated by block groups, is my dependent variable. 4

16 CHAPTER 2: LITERATURE REVIEW 2.1 The public transit situation and issues of the U.S and Richmond, VA City public transit systems play a vital role in carrying large shares of personal travel in metropolitan areas around the world. But as shown in Figure 2, in most places, public transit has been losing market share to private vehicles. In the market share of metropolitan travel, public transit is losing customers to private vehicles in the U.S. Nationally, only 2.1% of all trips were on public transit in 2001, compared to 85.8% by private vehicles, 9.9% by foot and bicycle, and 2.2% by other means (U.S. Department of Transportation, 2001). Transit use is higher in the centers and downtowns of the bigger city and metropolitan areas. In the U.S., New York City is the 800-pound transit gorilla nearly 4 in 10 (38%) transit trips nationally in 2000 were made in the greater New York City area (American Public Transportation Association, 2001). 5

17 Figure 2: Annual transit ridership and annual ridership per capita in the US [Source: American Public Transportation Association (2001)] As seen in Figure 3, in recent years, with the economy recovering and employment rate improving and higher gas prices, American transit use has gradually been rising. From 1995 to 2009, public transportation ridership in the U.S. grew at a rate of 34 percent twice as fast as the population growth rate and 10 percentage points faster than the growth rate of vehicle miles traveled on our road and highway system (Amalgamated Transit Union, n.d.). According to a recent report released in 2012 by the American Public Transportation Association (APTA), Americans took 10.4 billion trips on public 6

18 transportation in 2011, the second highest annual ridership since 1957, and this was the sixth year in a row that more than 10 billion trips were taken on public transportation systems nationwide. Figure 3: Annual Transit Ridership. [Source: American Public Transportation Association (APTA), 2011] Conversely, while transit share has grown every year from 2000 to 2010, it cannot meet the need seen in the huge demand for public transit. And as Figure 4 indicates, its share of total travel continues to fall. Data from the U.S. Census and Nationwide Personal Transportation Survey (NPTS) support that conclusion. 7

19 Figure 4: Transit Mode Share. [Source: American Public Transportation Association (APTA), 2011] The Greater Richmond Transit Company (GRTC) is the primary public transportation provider in Richmond, VA. It operates a hub-and-spoke system and covers the downtown area near the Virginia Commonwealth University (VCU) Health campus and government buildings along Broad Street, but provides services only to Richmond City, parts of Henrico County and Chesterfield County (Figure 5). 8

20 Figure 5: Richmond Greater Richmond Transit Company (GRTC) Bus Route Map. Comprehensive Operations Analysis (2008) (Source: GRTC, 2008) Some main corridors and high transit demand areas still lack transit service and require transport infrastructure and transport routes (Figure 6). As Blumenberg and Shike note lack of fair and appropriate transport accessibility might result in a spatial mismatch between social groups and social benefits (Blumenberg and Shiki, 2003). 9

21 Figure 6: Bus Service Areas and Population Density. Comprehensive Operations Analysis (2008) (Source: GRTC, 2008) Because of these issues, the Richmond metropolitan area ranks No. 95 out of the 100 metropolitan areas in the U.S. in terms of share of working-age residents with access to transit (Tomer et al., 2011) (Figure 7). To relieve and solve these problems, GRTC and Richmond s local transit agencies completed transit plans, such as Transit Development Plan and Comprehensive Operations Analysis (COA), to provide a series of recommendations and plans to improve and optimize the existing bus routes and build a transfer hub. Richmond Regional Planning District Commission (RRPDC) made its final 10

22 technical report of the Richmond Regional Mass Transit Study (RRMTS) and cooperated with VCU to research the four main transportation corridors of Richmond. Figure 7: Combined Ranking of Access to Transit and Employment, 100 Metropolitan Areas (Source: Tomer et al., 2011) However, the studies from Richmond s local transit and planning agencies were broadly written. I am going to identify the most vital variables and relative factors to analyze Richmond city s bus ridership to fill this void. This study researching city transportation is valuable to Richmond City development and policy modification. Because the major objectives of urban transportation policy are the achievement of sound land use patterns, 11

23 the assurance of transportation facilities for all segments of the population, the improvement of overall traffic flow, and the meeting of total transportation needs at minimum cost. Only a balanced transportation system can attain these goals - and in many urban areas this means an extensive mass transportation network fully integrated with the highway and street system (U.S. Congress, Senate, 1962). 2.2 Previous study of factors impacting transit ridership Factors such as population and employment distribution and density, service, fare, work locations and hours, number of automobiles and transit waiting time influence transit ridership. European Commission on Transportation Research (ECTR) categorizes two groups of direct strategies and indirect strategies to distinguish the similar variables. Direct strategies influence transit ridership efficiently and effectively, as do as external factors. Based on the ECTR, direct strategies include fare, service quality, marketing, and facilities (Table 1 and Table 2). 12

24 Table 1: Direct Strategies. (Source: European Commission Transportation Research, 1996) Direct Strategies Pricing Service Service Quality Priority Measures Regulatory Regime Information Others Fare Levels Ticketing Regimes/Fare Structure Ticketing Technology Subsidy Regime Extensiveness of Routes Distance to/from Stops Service Frequency/Travel Time Operating Hours Fleet Size Vehicle Characteristics Bus/Rail Stop Quality Interchange Quality Quality/Number of Staff Link Priority/Right-of-Way Junction Priority Market Regulation Operational Regulations Quality Regulations Information Provision Publicity/Promotion Park-and-Ride Integrated Approach 13

25 Table 2: Indirect Strategies (Source: European Commission Transportation Research, 1996) Indirect Strategies Car Ownership Car Use and Area- Specific Car Use and General Others Taxation of Car Ownership Restrictions on Car Ownership Traffic Calming Access Restrictions Road Pricing Parking Availability Cost of Parking Parking Enforcement Fuel Tax Restrictions on Car Use Car Vehicle Specification Information on Traffic Conditions Land use Planning Telecommuting/Tele- Shopping Flexible Working Hours Increase in Road Capacity Improvements to Non- Motorized Modes Taylor and Fink (2002) state transit ridership factors can be categorized into two groups: traveler attitudes and perceptions, and environmental system, and behavioral characteristics. The category of environmental system and behavioral characteristics, includes aggregated and disaggregated studies like the unit of research and metropolitan and big cities variables and individual traveler choice decisions. They also say the research of travelers and operators attitudes and perceptions are descriptive analyses 14

26 because transit operators often develop descriptive analyses for marketing and fare policy. But the research of environments, systems and behaviors are causal analyses (Figure 8). Descriptive and causal analyses each have advantages and disadvantages. Descriptive analyses are based on sets of often interesting and rich qualitative data from surveys of and interviews with transit operator staff. Thus, these studies focus on what transit managers believe affect transit ridership (Taylor and McCullough, 1998). Figure 8: Studies of Transit Ridership. (Source: Taylor and Fink, 2002) 15

27 2.3 Internal and external factors What I talk about the descriptive and causal analyses can be categorized into external factors and internal factors. Although they are named external factors and internal factors, actually they have a big relevance and affect each other. For instance, as a result of an increased population or employment density in the study area, the transit services also change. And if ticket fares are reduced, ridership will grow and the transit services also change. Both internal and external factors play a significant role in influencing transit ridership. Internal factors are things like public transit service, fare structures, bus route design and schedules, bus size, and policies. External factors are population and employment situation, job distribution, traffic congestion, parking distribution and costs and gas price. Public transit ridership is influenced by a variety of factors, both internal and external to the transit system. Internal factors are those under the purview of transit managers and policy boards, such as the level of service provided, fare structures and levels, service frequency and schedules, route design, and service area size. External factors, in contrast, are those outside of a transit agency s control-such as population and employment growth, residential and workplace location-and factors that influence the relative attractiveness of transit, such as gasoline prices and parking costs (Mineta Transportation Institute, 2002). 16

28 Dajani and Sullivan (1976) use a causal model to estimate public transit ridership with 1970 census data. The variables include median household income, density, transit service quality, percentage of downtown city workers, percentage of African-American population and auto ownership. Ben-Akiva and Lerman (1985) find that waiting time, especially at a stop or station impacts the transit ridership. Similarly, People don t mind waiting for a bus if they know how long it s going to be. Even if they have to waste the time, at least they know it s going to be 15 minutes. Otherwise they re sitting there thinking the bus will be along in about two minutes, and when it doesn t show, then they start getting frustrated (Duffy, 2002). Cervero (1990) states when considering the relation between transit service and riders and between ticket fare and riders, riders pay more attention to public transit service. So the public transit service is more important than ticket fare. That is valuable to transit agents and operations. McLeod, Flannelly and Behnke (1991) use data for the range from 1956 to 1984 of Honolulu, Hawaii to calculate multivariate variables time series regression models. In their models, using revenue trips, have five independent variables: kinds of jobs, adjusted per capita income, fares, and different size of the transport and a series of accounting by different disruptions. They find the gasoline price has a small impact on public transit ridership. 17

29 McLeod, Flannelly and Behnke (1991) find number of tourists, gasoline prices and free ticket riders are not important factors influencing public transit ridership. They use the number of passenger vehicles and other variables in two time-series regression models. Liu (1993) builds a regression model based on the data of Portland, Oregon. He uses the model to examine and evaluate per capita transit trips. The factors in his model are per capita transit capacity, per capita passenger car registrations, per capita transit subsidies, per capita income, percent of population residing in the central city, metropolitan area population, motor vehicle fuel prices, a time-trend variable for a period , annual total transit miles, average passenger fare, total employment in the Portland metropolitan area, and the effects of World War II. His results that show per capita income and auto ownership are important. Liu (1993) and Kain and Liu (1995, 1996) use regional employment as variables in their regression analyses. Chung (1997) states employment and regional development play a vital role in the Chicago Transit Authority (CTA) system. The female labor force is increasing, more and more women use private vehicles rather than public transit, but women were relying on public transit in the past (Rosenbloom and Burns, 1993; Hayghe, 1996). 18

30 Car ownership has a huge impact on improving public transit ridership. High-income passengers have cars and low-income passengers without cars are relying on the public transit (Kain and Liu, 1995). The effect of income has similarities with car ownership. Nelson and Nygaard (1995) find in all of the 40 land use and demographic variables, housing and employment density per acre are most important in transit demand. They can explain 93 percent of the variation in transit demand. Kain and Liu (1995) state the average ticket fares, revenue vehicle miles of service, regional employment levels and car ownership situation have huge impacts on public transit ridership. In his regression models, Gomez-Ibanez (1996) used both internal (ticket fare and transit service policies) and external (employee income, population) factors to study their influences on transit ridership and deficiency reduction for the Massachusetts Bay Transportation Authority (MBTA) in Boston. His models find external factors play a vital role in Boston. For instance, the ratio of downtown jobs and the percentage increase in per capita income affect public ridership a lot. Conversely, the transit service level and the fare influence can be ignored in Boston. Kain and Liu (1996) research and analyze the main factors which influence the level of public transit ridership based on the data for 184 systems from 1960 to 1990, 30-year long range. They use regression models to study , and ridership factors. In their regression model, they use the independent variables both in 19

31 public and private system such as revenue miles of service supplied, population, employment, population density, fraction of carless households, fare levels. And the models show ridership changes between 1980 and 1990 had R 2 = 0.75 or above. For the race, Blacks and Hispanics are more dependent on public transit (Pisarski, 1996; Rosenbloom, 1998). In this study I choose the Black rate as an independent variable, because there is a high proportion of Blacks in Richmond City. Previous literature shows low-income households and households without access to vehicles depend on public transit and previous research also documents how age, ethnicity and gender influence public transit ridership. Income is a factor to influence public transit ridership. More middle-income and highincome people don t choose public transit, but low-income passengers are increasing (Pucher et al., 1998). Based on the data of five cities (Seattle, Portland, Salt Lake City, Denver, and San Diego), Spillar and Rutherford research the relationship between city resident density and public transit ridership and the relationship between income and public transit ridership. They use total population, annual income situation and research area acre information to find out that density has a huge impact on transit ridership in the low-income area, but has little effect on the high-income group in public transit ridership (Spillar and Rutherford, 1998). Based on data of 85 Canadian city transit agencies from 1992 to 1998, Kohn (2000) examines the vital explanatory variable to forecast public transit ridership and then states 20

32 that average fares and revenue vehicle hours are the two key variables. In his model, he analyzes demographics, hours of transit service, fare structure, vehicle statistics, energy consumption, employment situation, passenger statistics, and revenue. Two variables, average fares and revenue vehicle hours explain all variation in the public transit ridership (R 2 = 0.97); other variables are meaningless. In his research, Kohn first picks up average fare in the regression model but R square shows this is not important. When he chooses population of the research area, the R square value increases but it is still low. The second step shows the population of the research area is more important than fare. He picked these two in one model and the R square value is And then more independent variables were added in his model to test the relationship: For each year of data (to account for any differences on an annual basis), for cities that have populations in excess of one million, this dummy variable assumed that larger cities have more comprehensive transit systems, more traffic, a greater dispersion of people geographically, longer commute times, and, perhaps, a greater tendency towards transit ridership, for cities with more than one million urban transit passengers; this dummy variable was similar to the preceding variable for cities with populations in excess of one million; despite the apparent similarity, the correlation between the 2 variables was only 0.19, for cities with populations less than 100,000; this dummy variable assumed that cities with populations less than 100,000 people have less comprehensive transit systems, less population dispersion, shorter commute times and, perhaps, a lower tendency to use public transit than cities with greater populations (Kohn, 2000). 21

33 In his survey, Syed (2000) finds that fare is the least influential factor comparing with transit information, customer and street service, station and on-board safety in passenger considerations. Kikuchi and Miljkovic (2001) consider the demographic conditions around the bus stops, conditions of the bus stop and the level of transit service to build and research the public transit ridership prediction model by bus stops. Also, Florida Department of Transportation (FDOT) uses t-test in transit ridership evaluation model to evaluate the public transit ridership by route, bus route direction and bus stop buffer characteristics, etc. (FDOT 2004; FDOT 2005). Residential and employment densities are critical determinants of transit ridership (Taylor and Fink, 2002). And Pushkarev and Zupan (1977) find that there is a positive influence between density and public transit ridership. Chu (2004) builds a public transit ridership model at the bus stop to study an average weekday boarding with six categories of factors like socio-demographics in the area, Transit level-of-service (TLOS) value, street environment for pedestrians, accessibility to population and employment and competition with other TLOS stops. TLOS based on transit availability and mobility and demographic characteristics, pedestrian environment, interactions with other modes, and competition from other bus stops were considered and found to play a significant role in predicting the ridership (Srinivas and Mahesh, 2012) 22

34 And most exterior factors are socioeconomic, and while there is no obvious line to distinguish interior and exterior factors, external factors are more important than interior factors. Although a wide array of factors clearly influence transit patronage, our analysis finds that the most significant factors influencing transit use are external to transit systems (Mineta Transportation Institute, 2002). Chen and Suen (2010) use the data of the year 2000 Census Transportation Planning Package (CTPP) to analyze and estimate production-side and attraction-side transit ridership by transportation analysis zones (TAZs) in Richmond. And they use both internal and external factors which affect transit demand as independent variables. As a result, they find bus stops per worker, auto density and population density are three most important factors in production-side analysis (Table 3). In the attraction-side analysis, percentage of zero-vehicle workers, and percentage of workers whose households are below poverty status level are key factors (Table 4). Table 3: Regression Model Parameter Estimates in Production Side (Source: (Chen and Suen, 2010)) Model B Std. Error Beta t Sig. (Constant) Bus stop/worker Automobile density Population density Percentage of the workers whose households are below poverty status

35 Percentage of the disabled workers Percentage of the senior workers R square = Table 4: Regression Model Parameter Estimates in Attraction Side (Source: (Chen and Suen, 2010)) Model B Std. Error Beta t Sig. (Constant) Percentage of the workers whose households have zero vehicles Percentage of the workers whose households are below poverty status Percentage of the workers making trips during a.m. and p.m. peak periods. Percentage of the disabled workers Bus stop/worker R square =

36 In summary, most of the literature researches both internal and external factors, such as household income, density, transit service quality, vehicles, fare, employment and population. This study will examine public transit ridership using these factors in Richmond City and also add gender and bus lines as new independent factors which they have not been used in past research. 25

37 CHAPTER 3: METHODOLOGY 3.1 Method and data collection This study will be conducted based on literature reviews and a rigorous data analysis on the transit ridership influenced by transit and group-level socioeconomic variables. In this study, geographic information system (GIS) is used to map factor indication and bus stop distribution by block groups and Statistic Correlation and Regression model to analyze the main factors (Total population, Households, Race, Bus stops count, etc.) which influence public transit ridership significantly. The principal data source is American Community Survey 5-Year Estimates of Richmond City and 2000 Census Transportation Planning Package (CTPP). And the dependent and independent variables are all from American Community Survey 5-Year Estimates of Richmond City. According to the 2007 household survey conducted by GRTC (2008), 86 percent of fixed route service is provided to downtown area and most local bus routes exist and support downtown area. More than 80 percent of the ridership was made up by local bus routes, so the research is concentrated on Richmond City. 26

38 3.2 Why use these methods Regression analysis is a statistical tool for the investigation of relationships between variables. Researchers collect data on the underlying variables of interest and employ regression to estimate the quantitative effect of the causal variables upon the variable that they influence. GIS is used to show the spatial distribution of each socioeconomic variable by block group level to display the different distribution. Many researchers use these methods to analyze the relationship between independent variables and dependent variables. Therefore, these two methods are used to analyze the relationship between one dependent variable and 14 independent variables. 3.3 Define variables Public transit provides a convenient and low-cost mode for residents. In this study, the dependent variable is the number of passengers boarding at the bus stop. Bus stops are typically located according to a local transit agency s service decisions and standards. For the independent variables, comparing with the strategies from transit agencies, macroeconomic conditions like socioeconomic factors influence public transit ridership more than internal factors. Most external factors are socioeconomic factors. No obvious line separates internal and the external factors, but of the two external factors are more important. 27

39 Therefore, based on literature review, independent variables used in previous literature not directly related to public transit ridership are excluded from this study. External factors like population (total population), population density (total persons/acre),employment density (employees/acre), income (median household income) car ownership (the number of vehicles in household), acres (the size of each block group), black ratio (black population/total population), female rate (total female population/ total population),households (total households), household density (households/acre), low-skilled jobs (total low-skilled Employment), low educated people (educational attainment under and not including high school), unemployment rate (total unemployed/total population), and the number of bus lines in each block group are chosen as independent variables. Table 5 shows both dependent and independent variables. Table 5: Dependent and Independent Variables Variable Type Variable Name Variable Definition Dependent Variable Daily-On The number of daily number of passengers boarding on bus at any bus stop by block groups Independent Variable TotPop Total population Income Vehicles Popden Huden Black ratio Median household income The number of vehicles in household Population density: persons/acre Household density: number of housing units per acre, households/acre Black ratio: Black population/total population(all races) 28

40 Acres Female rate LowEduction Households EmpDen LowSkilled Unemployed Rate NoBusLines The size of each block group Total female population/ Total population Educational attainment (under and not including high school) Total households Employment Density: employees/acre Total low-skilled Employment(jobs) Total unemployed/ Total population The number of bus lines 2007 NAICS Codes: Low skilled job (Construction, Manufacturing, Wholesale Trade, Retail, Transportation and Warehousing) 3.4 Hypothesis The hypothesis considers the relation between each independent variable (total population and density, acres, household density, black ratio, the number of bus lines income, vehicle count, low-skilled employment and employment density) and the dependent variable (public transit ridership) (Table 6). 29

41 Table 6: Hypothesis of independent factors Positive factors Negative factors Total population and Population density Unemployment rate Acres Household and Household density Bus line count Low education Female rate Black ratio Low-skilled employment Employment density Income Number of vehicles In the hypothesis, total population, population density, unemployment rate, acres, household and household density, bus line count, low education, female rate, black ratio, low-skilled employment and employment density are assumed to have positive impacts on the number of daily number of passengers boarding a bus at any bus stop by block groups. Income and number of vehicles have negative impacts on the dependent variable. 30

42 CHAPTER 4: ANALYSIS 4.1 Mapping dependent and independent variables For the group of variables utilized in the statistics model in this study, first they are generated by GIS to display in the maps to show each variable situation and distribution. Because of the data limitation, I don t collect all the block groups data and there are no data in the blank block groups. Following are the dependent and independent variables which are manipulated by GIS. Daily-On (Figure 9), Total population (Figure 10), Income (Figure 11), Car ownership (Figure 12), Population density (Figure 13), Household density (Figure 14), Black ratio (Figure 15), Acres (Figure 16), Bus line count(figure 17), Female rate(figure 18), Low education population (Figure 19), Households (Figure 20), Employment density (Figure 21), Low-skilled jobs (Figure 22), Unemployed rate(figure 23). 31

43 Figure 9: Daily-On Figure 10: Total population Some downtown areas are the most concentrated areas of daily-on boarding. Population is mainly distributed in the west and south of Richmond city and some downtown areas. 32

44 Figure 11: Income Figure 12: Vehicles High-income population is mainly distributed in the northwest and south of Richmond city. Some downtown areas, the south and west of Richmond city have a large number of private cars. 33

45 Figure 13: Population Density Figure 14: Household Density Some downtown areas and northeast of Richmond city have much greater population density. Some downtown areas and northeast of Richmond city have much greater household density. 34

46 Figure 15: Black Ratio Figure 16: Acres A larger proportion of blacks are in the northeast and south of Richmond city. Blocks in the central, southeast and west of Richmond city are much larger. 35

47 Figure 17: Bus line count Figure 18: Female Rate Bus lines are mainly located in the northwest of Richmond city and downtown areas. There is not much difference of distribution of female rate in each block of Richmond city. 36

48 Figure 19: Low Education Population Figure 20: Households Low-education population is mainly distributed in the south and east of Richmond city. Households are mainly distributed in the central and southwest of Richmond city. 37

49 Figure 21: Employment Density Figure 22: Low-Skilled Employment Some downtown areas, some north and south of Richmond city have greater employment density. Low-skilled jobs are distributed in the north and southeast of Richmond city and some downtown areas. 38

50 Figure 23: Unemployment Rate Some areas of north and south of Richmond city and some downtown areas have larger proportion of unemployed. 39

51 4.2 Statistics Model analysis In order to test the hypothesis, based on the correlation model in the Appendix A, Appendix B and Appendix C, the relationships are as follows: (1) Total population, acres, employment density, low-skilled employment and bus line count are positive with the dependent variable which are consistent with the hypothesis. (2) Number of vehicle is positive with the dependent variable which is against the hypothesis. (3) Income is negative with the dependent variable which is consistent with the hypothesis. (4) Population density, household density, black ratio, female rate, low education, households and unemployment rate are negative with the dependent variable which are against the hypothesis. In the beginning, I ran the variables in both ways (original and log transformed). But the results show almost no difference between the two ways and the R square is a little bit lower in the log transformed model. So finally I decided not to use log transformed variables. Only original variables are used in my statistical analysis. In the correlation model summary in the Appendix A, Appendix B and Appendix C, based on the Pearson Correlation and Sig, I exclude two independent variables (Popden, Households) which can be represented by Huden and Vehicles. 40

52 In the regression model, there are one dependent variable and twelve independent variables entered, but the stepwise regression model excludes 10 less significant independent variables and there are 2 most important independent variables retained, NoBusLines and Income (Table 7). Table 7: Regression model summary and parameter estimates Variables Entered/Removed a Model Variables Entered Variables Removed Method 1 Stepwise (Criteria: Probability-of-F-toenter <=.050, NoBusLines. Probability-of-F-toremove >=.100). 2 Stepwise (Criteria: Probability-of-F-toenter <=.050, Income. Probability-of-F-toremove >=.100). Dependent Variable: Daily_On Model Summary Std. Error of the Change Statistics Model R R Square Adjusted R Square Estimate R Square Change F Change a b Model Summary Change Statistics Model df1 df2 Sig. F Change 41

53 a. Predictors: (Constant), NoBusLines Predictors: (Constant), NoBusLines, Income ANOVA a Model Sum of Squares df Mean Square F Sig. 1 Regression b Residual Total Regression c Residual Total a. Dependent Variable: Daily_On b. Predictors: (Constant), NoBusLines Predictors: (Constant), NoBusLines, Income Table 8: Regression Model Parameter Estimates Coefficients a Standardized Unstandardized Coefficients Coefficients Model B Std. Error Beta t Sig. 1 (Constant) NoBusLines (Constant)

54 NoBusLines Income Dependent Variable: Daily_On Excluded Variables a Collinearity Statistics Model Beta In t Sig. Partial Correlation Tolerance 1 TotalPop b Income b Vehicles b HUDEN.033 b Acres.001 b FemaleRate b LowEdu b EMPDEN b UnempRate.070 b BlackRate.096 b LowSkilled.006 b TotalPop c Vehicles c HUDEN.022 c Acres.020 c FemaleRate c LowEdu c EMPDEN c

55 UnempRate.017 c BlackRate.022 c LowSkilled c a. Dependent Variable: Daily_On b. Predictors in the Model: (Constant), NoBusLines c. Predictors in the Model: (Constant), NoBusLines, Income In this regression model, it depicts statistically significant relationships between independent variables and daily-on dependent variable to estimate the different importance of key factors. Overall, based on the stepwise regression model, it can be determined that NoBusLines (variable name: the number of bus lines) and Income (variable name: Median household income) are the most vital factors impacting the amount of daily number of passengers boarding on bus at any bus stop by block groups. More bus lines is significantly associated with higher levels of bus transit ridership at any bus stop. Conversely, income is negatively related to bus transit ridership. Table 9: Summary of key impacting variables Independent Variables Variable name NoBusLines Income The number of bus lines Median household income 44

56 4.3 Model Validation The above stepwise regression model has yielded the following estimated equation: Daily_On = *NoBusLines-0.003*Income For the purpose of model validation, the input data for variables NoBusLines and Income of 159 census block groups [Note: two block groups (GID# , and GID# ) are excluded due to their missing income data] are plugged into the above equation, which yields the modeled Daily_On (reflecting average trend of bus boarding across board) against which the observed Daily_On (reflecting actual bus boarding, which fluctuates around the average trend line) will be compared, and their percentage error will be calculated. In its Title VI report, GRTC uses 25% (±) as a benchmark to judge if a major service change has occurred. Following the same benchmark, this thesis also assumes that, for a block group: If the percentage error between observed Daily_On and modeled Daily_On is within 25% (±), the block group is adequately served by bus services, more or less reflecting average trend of bus boarding; 45

57 If the percentage error between observed Daily_On and modeled Daily_On is greater than + 25% (above the average trend line), the block group is overserved by bus services. To return to the normal, average trend, the number of bus lines and bus services should be decreased; and On the contrary, if the percentage error between observed Daily_On and modeled Daily_On is less than -25% (below the average trend line), the block group is underserved by bus services. Therefore, the number of bus lines and bus services should be increased in order to return to the normal, average trend. It should be noted the above determination of overserved/underserved status is not from the strict sense of transit supply and transit demand as defined in the so-called transit desert analysis. Instead, it merely compares the observed stop-level bus boarding against the trend line (reflecting the theoretical and average stop-level bus boarding). If GRTC wants to match the observed stop-level bus boarding to the trend line, its service levels and structures need to be properly adjusted. 46

58 Figure 24: Percent Error by Block Group in Richmond City Given the above assumption, Figures 24 reveals the following: First, downtown Richmond and its immediate neighborhood are overserved by GRTC; Second, the west side and south side of the City are underserved by GRTC; and 47

59 Third, the rest of the City is more or less adequately served by GRTC. Following this line of thought, this thesis recommends that GRTC gradually change the existing hub-and-spoke bus system structure and move some bus services from downtown Richmond to the west side and south side of the City. 48

60 CHAPTER 5: CONCLUSION This study estimates the Richmond City public ridership based on a group of independent variables. With Richmond s high demand for bus transit, how to satisfy the improving bus transit is a challenge problem. Although Richmond GRTC well serves the downtown city area, urban fringe areas like Midlothian, South Side are not well covered and no bus lines cross some suburbs. The existing hub-and-spoke bus transit system needs to be refined and optimized to adapt future development. As the result, I find bus stop count and bus line count definitely impact public transit ridership. Government and local agencies should consider investing in suburb-to-suburb bus transit and optimizing the bus line locations and amount, bus stop locations and number of bus stops. The analyses presented in this thesis are particularly valuable for local and regional policy-makers to improve public transit ridership and to deal with transit and environmental problems like traffic congestion reduction, crime, civic engagement, the enhancement of economic development, gas conservation and air quality improvement. The evidence and resulting thesis also support GRTC and local planning seeking transportation funding to pursue balanced public transit future plans and land use plans. I recognize that, because of the data limitation, I did not collect all of the block group data. And some considerable factors like inflation-adjusted per capita income, revenue vehicle miles, fraction of carless house and fare structure in different years will be joined 49

61 in the future study. Multicollinearity cannot be ignored, because it leads to the result that the coefficient estimates are unstable and some variables difficult to interpret. 50

62 APPENDIX A Correlation model summary (All Variables) Daily_On Income Vehicles POPDEN HUDEN BlackRate Acres FemaleRate Daily_On Pearson Correlation Sig. (2-tailed) N Income Pearson Correlation ** ** ** * Sig. (2-tailed) N Vehicles Pearson Correlation ** * **.321 ** Sig. (2-tailed) N POPDEN Pearson Correlation ** ** * **.077 Sig. (2-tailed) N HUDEN Pearson Correlation *.842 ** ** ** Sig. (2-tailed) N BlackRate Pearson Correlation ** ** * ** ** Sig. (2-tailed) N Acres Pearson Correlation ** ** ** Sig. (2-tailed) N FemaleRate Pearson Correlation * ** Sig. (2-tailed) N LowEdu Pearson Correlation **.274 ** *.458 ** * Sig. (2-tailed) N